Better use Bayes: A Comparison of Classical and Bayesian Age-Models of Sedimentary Deposits by Maarten Blaauw

Loading Map....


Date(s) - 04/02/2019
4:00 pm - 5:30 pm

HS XVIIb Main Building UoC


by Prof. Dr. Maarten Blaauw
School of Natural and Built Environment, Queen’s University Belfast

Reliable chronologies are of key importance for fossil-based studies of past climate, environment and human activity. Only when put on a common time-scale can multiple studies be compared, and can spatio-temporal patterns of past events be properly identified and interpreted. However, producing chronologies is not a trivial task, especially if, as is too often the case, funds for dating are limited.

Many studies rely on ‘classical’ age-modelling techniques such as linear interpolation, to obtain age estimates for dated and undated core depths of sedimentary deposits. This method might be so popular because it often provides pleasingly narrow uncertainty estimates – in fact, the longer the distance between dated depths, the narrower the confidence intervals, so, fewer dates give you a more precise age-model! Recently developed Bayesian approaches produce age-depth models that aim to simulate the sedimentation process, using ‘random walks’ between dated depths that can be constrained by limits on variability in sedimentation. They can also deal with outlying dates.

Using a range of real-world and simulated dated cores, we apply classical as well as Bayesian age-models and compare the precision estimates and measures of accuracy, for cores dated at very low to very high resolution. Our analysis shows that classical age-models tend to be highly over-optimistic, whereas the Bayesian models produce realistic uncertainty estimates and are reliable and robust at an impressive range of dating densities as well as scatter and outliers.